JP4732153B2 - Pixel interpolation method and apparatus, and image reading apparatus - Google Patents

Description

  The present invention relates to a pixel interpolation method and apparatus used at the time of reading an image using a pixel reading element, and an image reading apparatus using the mechanism, and more particularly to a technique for giving a pixel level value of a pixel to be interpolated missing in an image. It is concerned.

  An image reading apparatus for reading a form image and performing image processing such as OCR is known. For example, the form reading device disclosed in Patent Document 1 by the present applicants reads image data of a form, stores an OCR memory that stores binarized data obtained by binarizing the image data, and reads the entire read image data. An overall image memory for storing, and performing character recognition from the binarized data stored in the OCR memory and outputting it, and outputting the image data stored in the overall image memory.

JP 2004-287927 A

  A contact sensor (contact image sensor, CIS) used in such a reading apparatus is mainly a system in which a plurality of sensor chips cut out from a wafer are arranged side by side. A large number of pixel reading elements are formed on the sensor chip. In order to further arrange image reading elements corresponding to the reading width, the sensor chips are arranged in parallel as shown in FIG.

  Although it is desirable that there is essentially no gap between the sensor chips, since it is physically difficult in practice, it is arranged to be as close as possible to one pixel or less. However, this method has a problem that gaps that are not imaged are generated between the sensor chips and the gap portions are crushed. When character recognition such as OCR is performed, the influence of this gap has a serious effect such as a reduction in recognition rate.

Therefore, as shown in FIG. 12, the sensor chips (120) and (121) are arranged so that the gap (122) between the sensor chip 1 (120) and the sensor chip 2 (121) is exactly one pixel width, and the gap A method of interpolating (122) based on peripheral pixel information is used.
As an interpolation method, techniques disclosed in Patent Documents 2 to 4 are known.

JP 2000-196835 A JP 2004-274573 A JP 2004-356933 A

Japanese Patent Application Laid-Open No. 2004-228561 prevents image information from being lost when reading a document due to a gap between sensor chips, measures gap information between sensor chips of a plurality of sensor chips, and outputs image signals from pixel reading elements. And a technique for interpolating an image signal between sensor chips based on the gap information.
While this technique achieves interpolation while suppressing the processing load, linear interpolation is used for interpolation, and image reproducibility is not sufficient.

Patent Document 3 discloses a technique for calculating interpolation pixel data for interpolating unreadable pixels (pixels of interest) between image reading sensors using a filter. A plurality of filters having different filter sizes are provided as filters, and the pixel interpolation unit detects the flatness of pixels in the vicinity of the target pixel, and switches the plurality of filters according to the detection result.
According to the present technology, in addition to detecting the flatness of a pixel in the vicinity of the pixel of interest, it is necessary to perform a process of switching the filter means, which may cause a complicated process and a high processing load. In addition, since interpolation is performed using several filters, optimal interpolation pixel data is not always obtained.

Patent Document 4 discloses a technique for interpolating using a plurality of pixel data on both sides of a missing pixel, thereby making the pixel data of the missing pixel more appropriate even when there is a density peak at the corresponding position of the missing pixel. Can be interpolated.
Certainly, according to this method, even a pattern that could not be reproduced sufficiently in the past may be able to be accurately interpolated to some extent, but many patterns other than those assumed by the disclosure are the same as in the past. There is no high reproducibility.

  As described above, the pixel interpolation method disclosed in the related art does not have sufficient reproducibility of the outline / inside of a character, which is indispensable particularly in the case of OCR of a form, and a reduction in recognition rate is inevitable. Alternatively, if an advanced estimation method is used to improve reproducibility, a great amount of computer resources are consumed, and high-speed and simple processing cannot be realized.

  The present invention was created in view of the above-described problems of the prior art, and is an accurate pixel as an interpolation target pixel missing in an image read by scanning a plurality of parallel pixel reading elements. A level value is given and a simple and high-speed process is realized.

In order to solve the above problems, the present invention provides the following means.
That is, according to the first aspect of the present invention, when an image is read by scanning the reading target with a plurality of pixel reading elements arranged in parallel, the pixel level value of the interpolation target pixel missing in the image is given. This is a pixel interpolation method.
In this method, first, the pixel level value group acquisition means is in a direction opposite to the first direction and the pixel level value group assigned to a predetermined number of pixels connected in a first direction from the interpolation target pixel. A pixel level value group acquisition step of acquiring a pixel level value group assigned to a predetermined number of pixels connected in a second direction;

Subsequent to the pixel level value group acquisition step, the pixel level prediction calculation means uses the pixel level prediction formulas expressed by the following formulas 10 to 12 for each pixel level value group in the first direction and the second direction. The pixel level prediction calculation step for calculating the respective pixel level prediction values FD_L and FD_R , and the weighting value calculation means, for each pixel level value group, the opposite side of the sum of the squares of the deviations in the first direction and the second direction. The weight value calculation for calculating the weight value in each of the first direction and the second direction using a weight calculation formula that performs weighting weighted in the first direction and the second direction by taking the ratio of the squares of the deviations Process the steps simultaneously or in any order.
(Equation 10)
When the level value of the pixel is P1> P2 and P4 <P5
Predicted value FD_L = P2- (| P1-P2 |) in the first direction
Prediction value FD_R in the second direction = P4- (| P4-P5 |)
(Equation 11)
When the level value of the pixel is P1 <P2 and P4> P5
Predicted value FD_L = P2 + (| P1-P2 |) in the first direction
Prediction value FD_R = P4 + (| P4-P5 |) in the second direction
(Equation 12)
When the level value of the pixel is other than the relationship of Equation 1 or Equation 2
Predicted value FD_L = P2 in the first direction
Prediction value FD_R = P4 in the second direction

  Further, the interpolation target pixel level value calculating means multiplies the pixel direction predicted value in the first direction by the weighting value in the first direction, and the pixel level predicted value in the second direction in the second direction. An interpolation target pixel level value calculating step of calculating a pixel level value of the interpolation target pixel using an interpolation target pixel level calculation formula including a calculation of multiplying the weighting value of.

  According to a second aspect of the present invention, in the pixel level prediction calculation step, at least two pixel level values are extracted from each of the pixel level value groups, and the pixel level values are in a first direction and a first direction. In any of the two directions, when rising toward the interpolation target pixel, or when falling toward the interpolation target pixel, it corresponds to the interpolation target pixel from the rising rate or falling rate in each direction. The pixel level value is calculated and used as a pixel level predicted value. Otherwise, the pixel level value of a pixel connected to the interpolation target pixel in each direction is used as the pixel level predicted value. Is.

According to a third aspect of the present invention, in the weighting value calculating step, a first average value obtained by averaging the pixel level value group in the first direction and a pixel level value group obtained by averaging the pixel level value group in the second direction. 2 An average value and a median average value obtained by averaging pixel level values given to a predetermined number of pixels connected in both directions from the interpolation target pixel are calculated.
Then, the first deviation between the first average value and the median average value and the second deviation between the second average value and the median average value are compared, and a larger weighting value is given to the smaller deviation. And

  According to a fourth aspect of the present invention, in the above-described pixel interpolation method, following the interpolation target pixel level value calculating step, the pixel level value limiting means has predetermined first and second directions respectively from the interpolation target pixel. A pixel level value restriction range based on at least one of the maximum value and the minimum value is determined from the peripheral pixel level value group assigned to the pixels of the peripheral pixel number, and the pixel level value of the interpolation target pixel obtained above is determined. It has a pixel level value limiting step for correcting to be within the range.

  The invention described in claim 5 provides a configuration in which at least one of a hue value, a saturation value, and a brightness value of a pixel can be used as a pixel level value in the pixel interpolation method.

The present invention can also be provided as a pixel interpolation device that implements the pixel interpolation method. That is, in the invention of claim 6, when an image is read by scanning the reading target with a plurality of pixel reading elements arranged in parallel, the pixel level value of the interpolation target pixel missing in the image is set. A pixel interpolation device is provided.
The apparatus includes a group of pixel level values assigned to a predetermined number of pixels connected in a first direction from the interpolation target pixel, and a predetermined value connected in a second direction opposite to the first direction. The pixel level value group acquisition means for acquiring the pixel level value group assigned to the number of pixels, and each pixel level value group using the pixel level prediction formula expressed by the following Expression 13 to Expression 15 Pixel level prediction calculation means for calculating the pixel level prediction values in the respective directions and the second direction.
(Equation 13)
When the level value of the pixel is P1> P2 and P4 <P5
Predicted value FD_L = P2- (| P1-P2 |) in the first direction
Prediction value FD_R in the second direction = P4- (| P4-P5 |)
(Equation 14)
When the level value of the pixel is P1 <P2 and P4> P5
Predicted value FD_L = P2 + (| P1-P2 |) in the first direction
Prediction value FD_R = P4 + (| P4-P5 |) in the second direction
(Equation 15)
When the level value of the pixel is other than the relationship of Equation 1 or Equation 2
Predicted value FD_L = P2 in the first direction
Prediction value FD_R = P4 in the second direction
Further, the weighting in which each pixel level value group is weighted in the first direction and the second direction by taking the ratio of the square of the deviation on the opposite side to the sum of the squares of the deviations in the first direction and the second direction. Weighting value calculation means for calculating the weighting values in the first direction and the second direction by inputting into the weighting calculation formula for performing the weighting, and the weighting in the first direction to the pixel level predicted value FD_L in the first direction And a pixel level value of the interpolation target pixel is calculated based on an interpolation target pixel level calculation formula including an operation of multiplying the predicted pixel level value FD_R in the second direction by the weight value in the second direction. Interpolation target pixel level value calculating means is provided.

  The pixel interpolation device according to claim 7, wherein a maximum value or a minimum value is obtained from a peripheral pixel level value group assigned to pixels having a predetermined number of peripheral pixels in each of the first direction and the second direction from the interpolation target pixel. A pixel level value limiting unit that determines a pixel level value limiting range based on at least one of the pixel level values and corrects the pixel level value of the interpolation target pixel calculated by the interpolation target pixel level value calculating unit to be within the range. Provide configuration.

The present invention further includes a pixel interpolation mechanism for reading an image by scanning a reading target with a plurality of pixel reading elements arranged in parallel in one column and for providing a pixel level value of an interpolation target pixel missing in the image. It can also be provided as an image reading apparatus.
The apparatus includes an image input processing unit that inputs pixel level values read by the pixel reading element in accordance with scanning and accumulates them as an image, and a pixel interpolation mechanism that includes means similar to the pixel interpolation device. In addition, an image output unit that outputs an image including an interpolation target pixel interpolated by the pixel interpolation mechanism is provided.

  The image reading apparatus according to claim 9, wherein the pixel interpolating mechanism is configured so that a maximum value is obtained from a peripheral pixel level value group assigned to pixels having a predetermined number of peripheral pixels in each of the first direction and the second direction from the interpolation target pixel. A pixel level value restriction that determines a pixel level value restriction range based on at least one of the value and the minimum value, and corrects the pixel level value of the interpolation target pixel calculated by the interpolation target pixel level value calculation means to be within the range A configuration comprising means is provided.

  According to the pixel interpolation method of the first to fifth aspects of the present invention, the pixel level value is predicted by a simple method from the pixel level values of the peripheral pixels connected to the missing interpolation target pixel, and the pixel level value is determined according to the direction. By predicting the pixel level value and weighting the predicted value, a more accurate prediction of the pixel level value is realized than in the prior art.

  The pixel interpolation apparatus according to claims 6 and 7 also enables high-speed and simple processing as described above and realizes accurate prediction of the pixel level value, so that it is suitable for input to an image processing apparatus such as OCR. A device for outputting a simple image can be provided.

  The image reading apparatus according to the eighth and ninth aspects of the present invention contributes to an image reading apparatus that can output a highly reproducible image by including a pixel interpolation mechanism.

Embodiments of the present invention will be described below with reference to the accompanying drawings. The present invention is not limited to the following examples and can be appropriately designed and implemented.
FIG. 1 is a processing flowchart of a pixel interpolation method according to the present invention. FIG. 2 is a block diagram of the image reading apparatus according to the present invention. In the present embodiment, in order to read an image for each pixel, a plurality of pixel reading elements (hereinafter referred to as a reading unit) arranged in a row are scanned on the original paper surface, and are acquired and input as electric signals by photoelectric conversion. .

The acquired pixel column includes color information as a pixel level value, and may be acquired as information such as a hue value indicating hue, a saturation value indicating the degree of color, and a brightness value indicating brightness. . Among these, in the present embodiment, red (R), green (G), and blue (B) are acquired in 256 steps from 0 to 255 as hue values. Accordingly, as the pixel level value, 256 levels of values are assigned to each RGB for each pixel.
The color image (1) thus obtained is input.

  The image (1) includes missing pixels due to the gap (92) between the sensor chips as shown in FIG. 12, and this position is known in advance from the structure information of the reading unit. Therefore, a pixel level value to be interpolated is assigned to the missing pixel by the following process.

  In this embodiment, when the scanning unit is scanned to interpolate the missing pixels in the gap between the sensor chips, pixels are continuously missing in the scanning direction. However, the present invention is not limited to this, and pixel reading is not limited thereto. It may be used to interpolate a defective pixel that cannot be read normally even if there is an element, or a defective pixel that cannot be read due to a failure of the element.

  First, a pixel interpolation mechanism that is a core part of the image reading apparatus of the present invention will be described. The pixel interpolation mechanism has a configuration shown in FIG. 2, a reading unit, a well-known input processing unit that primarily processes a signal from the reading unit such as analog / digital conversion, and an image output for outputting an image including an interpolation target pixel described later. It is a mechanism excluding the part.

  The input color image (1) is first processed in a pixel level value acquisition unit (20) which is a pixel level value group acquisition means. At this time, as shown in FIG. 1, the process is divided into a step of acquiring the pixel level value after conversion to gray and a step of acquiring the pixel level value as it is in color. The pixel level value acquisition unit (20) includes a gray conversion circuit (201) and a gray delay unit (202) as the former processing unit, and a color delay unit (203) as the latter processing unit.

First, the processing from the gray conversion step (S1) to the weighting calculation step (S5) in FIG. 1 will be described in order.
The gray conversion circuit (201) calculates a pixel gray level value from the input RGB pixel level value according to the equation ( 16 ) and converts it to a gray image. (S1)
(Equation 16 )
Gray = 0.3 x R + 0.59 x G + 0.11 x B

  Then, the gray image is delayed in five stages in the gray delay unit (202). That is, as shown in FIG. 3, two pixels connected from a missing pixel in a first direction (here, leftward) are extracted, and further in a second direction (rightward) opposite to the first direction. Two connected pixels are extracted, and five pixels including missing pixels are given different delays to the four pixels so that calculation processing can be performed.

Here, the pixel gray level value of the second pixel on the left side is acquired as Gr1, the value of the adjacent pixel on the left side is Gr2, the value of the adjacent pixel on the right side is Gr4, and the value of the second pixel on the right side is acquired as Gr5. To do. (S2)
When predicting the pixel level value of the missing pixel from the left and right pixel level values, the present invention makes an accurate prediction using the weighting value. The weight value is calculated by a weight value calculator (22) which is a weight value calculator.

The weight value calculator (22) is provided with an average value calculation circuit (221), a difference absolute value calculation circuit (222), and a weight calculation calculator (223).
First, each calculation process of Expressions 17 to 19 is performed by an average value calculation circuit (221) that calculates an average value of a plurality of pixel gray level values.

(Equation 17 )
Average value of gray density of left two pixels Average_LEFT = (Gr1 + Gr2) / 2
(Equation 18 )
Average value of gray density of right two pixels Average_RIGHT = (Gr4 + Gr5) / 2
(Equation 19 )
Average gray density of two neighboring pixels excluding missing pixels Average_CENTER = (Gr2 + Gr4) / 2

That is, in Equation 17 , the average value of the pixel gray level values of the left two pixels (30) is calculated, in Equation 18 , the average value of the pixel gray level values of the right two pixels (31) is calculated, and in Equation 19 , the missing pixels are the center. Average values of the pixel gray level values of the left and right pixels (32) are respectively obtained and output as Average_LEFT, Average_RIGHT, and Average_CENTER.
This process is an average value calculation step (S3) in FIG.

In the above description, the pixel level values of two connected pixels are averaged for the left and right average values, but an average value of a larger number of pixels, for example, an average value of about three pixels or four pixels may be used. The average value at the center is not limited to one pixel connected to the missing pixel, and may be an average value of about 2 to 3 pixels.
By using the average value in this way, it is possible to prevent processing with an inappropriate gray density even when a pixel connected to a missing pixel accidentally takes a peculiar value.

Next, the absolute difference calculation circuit (222) calculates the absolute value of the difference between the average value of the left and right pixels and the average value of the center pixel. (Step S4)
Equations ( 20) and (21) are the calculation formulas, and are used as indicators for measuring how large the change in gray density between the left and right pixel columns is.

(Equation 20 )
Left deviation (first deviation)
Difference_L = | Average_CENTER-Average_L |
(Equation 21 )
Right deviation (second deviation)
Difference_R = | Average_CENTER-Average_R |

From the above two deviations, the weight calculation calculator (223) performs an operation so as to give a larger weight value to the one with the smaller deviation. Here, the calculation method can be arbitrarily set, but in the present embodiment, the following equations 22 and 23 are used.

(Equation 22 )
Square of each deviation Diff_L2 = Difference_L * Difference_L
Diff_R2 = Difference_R * Difference_R
(Equation 23 )
Weighting value Weight_L = Diff_R2 / (Diff_L2 + Diff_R2)
Weight_R = 1−Weight_L

As is clear from Equation 22, by taking the ratio of the square of the deviation on the opposite side to the sum of the squares of each deviation, the larger the deviation on the opposite side (the smaller the deviation on the own side), the greater the weighting. Moreover, since a quadratic weighting curve is taken as shown in FIG. 4, weighting weighted to the left and right is performed. (Step S5)

As described above, in the weighting value calculation steps (S3 to S5), the pixel level value assigned to the left and right pixel columns is used to calculate the weighting value corresponding to the pixel level prediction value of the missing pixel described below. Can do.
It should be noted that the present invention proposes that the pixel level prediction value is weighted to perform accurate pixel interpolation, and the weighting method can be arbitrarily determined regardless of the above. In particular, the above formula proposes a suitable calculation formula, but a simple calculation formula or a more complicated statistical calculation may be used in accordance with the processing performance of the calculation circuit and the arithmetic unit.

On the other hand, in the color delay unit (203) of the pixel level value acquisition unit (20), the two left pixels of the missing pixel are processed from the input color image (1) by the same five-stage delay process as the gray delay unit (202). The pixel and the two pixels on the right are extracted, and the respective pixel level values are acquired. (Step S6)
At this time, since gray conversion is not performed, RGB hue values are obtained as pixel level values. Although the same processing is performed for each of RGB as the original processing, in the following, in order to simplify the description, the pixel level value and the pixel level predicted value are simply referred to generically, and the processing for each color is not mentioned.

  The pixel level values extracted by the color delay unit (203) are P1, P2, (missing pixels), P4, and P5 in order from the left. These values are input to a pixel level prediction calculator (21) which is a pixel level prediction calculator.

  The pixel level prediction computing unit (21) first compares the magnitude relationship between P1 and P2, and P4 and P5. Then, it is predicted what relationship the interpolated pixel to be interpolated has among the surrounding pixels. That is, the present invention assumes the following three patterns, two of which are shown in FIGS.

In FIG. 5, the pixel level value decreases in order from the left side, and increases in order on the right side. In such a pattern, the interpolation pixel is predicted to have a downward peak.
On the other hand, in FIG. 6, the pixel level value increases in order from the left side and decreases in the right side. In the case of such a pattern, it is predicted that the interpolation pixel has an upward peak.

Furthermore, there is a case where neither pattern is applicable. In accordance with each pattern, a pixel level prediction value to be given to the interpolation pixel based on the left and right pixel level values is calculated.
It is as few 24 for the pattern of first FIG.

(Equation 24 )
P1> P2 and when the P4 <P5 FD_L = P2- (| P1-P2 |)
FD_R = P4- (| P4-P5 |)

  This formula performs a proportional calculation based on the rate of decline from P1 (50) to P2 (51) to obtain the left predicted value FD_L (52), while the rate of decline is also obtained for P5 (53) to P4 (51). The right side predicted value FD_R (55) is obtained by performing a proportional calculation based on the above. Therefore, a predicted value can be calculated by proportional calculation even in a configuration in which pixels having different distances are extracted from connected pixels, not between adjacent pixels having the same distance as in the above example.

Similarly, the pattern of FIG. 6 is calculated based on Equation 25 .
(Equation 25 )
P1 <P2 and P4> when the P5 FD_L = P2 + (| P1 -P2 |)
FD_R = P4 + (| P4-P5 |)

Further, when neither of the patterns shown in FIGS. 5 and 6 is applicable, the pixel level values P2 and P4 of the pixels adjacent to the missing pixel in the left and right are set as the pixel level predicted values. That is, as shown in Equation 26 .
(Equation 26 )
When other than the above FD_L = P2
FD_R = P4

The pixel level prediction computing unit (21) of the present invention outputs a pixel level prediction value based on the pixel level value in each direction from the missing pixel according to the above pixel level prediction formula. (Step S7)
The steps (S1 to S5) centering on the above-described weighting value calculation step and the steps (S6 to S7) centering on the pixel level prediction calculation step can be performed in parallel at the same time. When the arithmetic processing is performed by one central arithmetic processor (CPU) or the like, the processing may be performed in any order.

Then, a weighted average process (S8) is performed on the predicted pixel level value obtained in the pixel level prediction calculation step using the weight value obtained in the weight value calculation step.
That is, in the weighted average calculator (23), which is the interpolation target pixel level value calculation means, the pixel level prediction value on each side is multiplied by the weight value, and the calculation of the number 27 is performed to add the results.

(Equation 27 )
Pixel level value to be given to the interpolation pixel A3 = (FD_L * Weight_L) + (FD_R * Weight_R)

  Such weighting processing has a remarkable effect as described below. That is, as shown in FIG. 7, when the rate of decrease from P1 to P2 is different from the rate of decrease from P4 to P5, and is gradually gradual, a pixel that is desirable in the conventional linear interpolation method that simply averages The value becomes larger than the level value, and appropriate interpolation cannot be realized. Similarly, in FIG. 8, the pixel level value is smaller than the appropriate value by linear interpolation.

  If an appropriate pixel level value cannot be obtained, pixels with an unnatural hue are arranged in a color image, which is not a preferable pixel interpolation. The same applies to a gray image, and particularly when binarizing from a gray image, the result differs depending on a slight difference in density, and therefore the necessity for performing optimal pixel interpolation is extremely high.

The processing of the present invention may be completed by the above weighted average processing (S8). However, as a more preferred embodiment, the following pixel level value limiting step can be added.
That is, the output range of the pixel level value obtained by the pixel level value limiting unit (24) which is a pixel level value limiting unit is limited. The pixel level value limiting unit (24) acquires a pixel level value of 8 pixels from the missing pixel to the left and right, respectively, and a density detection circuit (241) for detecting the maximum value and the minimum value, and the maximum value and the minimum value. An output density limiting circuit (242) for limiting the pixel level value so as to be within the value range is provided.

FIG. 9 is a diagram for explaining this processing.
When the pixel level values (density) of the left and right 8 pixels change as shown in the figure, the third pixel (90) point from the left takes the maximum density, and the sixth pixel point (91) from the right takes the minimum density. ing.
If the pixel level value assigned to the missing pixel is between the maximum density and the minimum density as indicated by the point (92), the pixel level value is output as it is.
On the other hand, if it exceeds the maximum density as indicated by the point (93), the pixel level value is output with the same value as the maximum density value. That is, the output is as shown by point (93).

Conversely, when the value is lower than the minimum density, the minimum density value is output as a pixel level value.
The purpose of this processing is that even if the pixel level value exceeds the maximum density in the above calculation, it may be extremely bright and inappropriate, so it is compared with the pixel level value of surrounding pixels. It is limited within the range.
As this application form, for example, a predetermined threshold value may be provided such that a certain increase is recognized based on the peripheral maximum density, and an upper limit (the lower limit is the same) may be set.

The output density limit circuit (242) outputs the pixel level value of the interpolated pixels within the limit range. This completes the operation of the pixel interpolation mechanism according to the present invention.
Further, in the image reading apparatus of the present embodiment, in order to insert the pixel level value obtained at the first missing pixel position, the interpolation position calculator (25), the original pixel position phase correction unit (26), the interpolation pixel output An image output unit composed of the unit (27) is provided.

  That is, the interpolation position calculator (25) counts the signal read by each pixel of the reading unit (CIS) by the system counter, detects that the first pixel at the left end of the reading unit has been read, and the pixel from that time. Start counting by the counter. As shown in FIG. 10, the system counter sequentially counts pixels, but the pixel counter counts for each sensor chip of the reading unit.

  For example, if 468 pixels are arranged in one sensor chip, it can be seen that the 469th pixel is a missing pixel to be interpolated in the present invention. Therefore, a signal indicating interpolation pixel position information is output from the interpolation position calculator (25) to the interpolation pixel output unit (27). (Step S11)

  On the other hand, the original pixel position phase correction unit (26) receives the color image (1) and delays the signal by an amount corresponding to the time required for performing the interpolation pixel calculation processing according to the present invention. For this purpose, as shown in FIG. 11, a process for synchronizing the output of the pixel level value of the missing pixel is performed using a delay line. (Step S12)

Finally, in the interpolated pixel output unit (27), the pixel level value of the missing pixel is inserted at the time of the interpolated position signal (flag) according to the interpolated pixel position information in step S11 while inputting the synchronized original image pixels. (Step S13).
By repeating this, all the missing pixels are interpolated in order, and can finally be output as a color image (2).

  As described above, the present invention can provide not only an image reading apparatus but also a pixel interpolation mechanism as a single apparatus. In addition, the configurations of the image input processing unit and the image output unit of the image reading apparatus can be appropriately designed in combination with an existing configuration in a known image reading apparatus.

FIG. 13 shows a comparison between a reading result obtained by the image reading apparatus and a reading result obtained by a conventional image reading apparatus.
When an image is read in a state where there is a gap, a blank corresponding to the gap portion is generated as in the upper character (130). According to the conventional interpolation method that does not perform weighting, there is a problem that interpolation is performed in a noise manner like the left character (131), which becomes an obstacle in subsequent processing such as OCR.
Since the method using weighting according to the present invention has the effects as shown in FIGS. 7 and 8, it is possible to obtain a more preferable reading result such as the right character (132).

  Since the method of the present invention can realize good pixel interpolation with the simple method as described above, it is meaningful to be mounted on a device that reads images at high speed and accurately, such as a device that reads forms. In addition to being provided as an image reading apparatus such as a scanner, it can also be used as a technique for interpolating the pixels of an accumulated image by implementing a pixel interpolation method in a computer that accumulates the read image.

4 is a processing flowchart of a pixel interpolation method (including an image reading method) according to the present invention. 1 is a configuration diagram of an image reading apparatus according to the present invention. It is explanatory drawing explaining the process of an average value calculation step. It is a graph curve of the weighting value in this invention. It is explanatory drawing explaining the calculation method of the pixel level estimated value by this invention. It is explanatory drawing explaining the calculation method of the pixel level estimated value by this invention. It is explanatory drawing explaining the effect of the weighting in this invention. It is explanatory drawing explaining the effect of the weighting in this invention. It is explanatory drawing of a process of an output density restriction | limiting step. It is explanatory drawing of the process by an interpolation position calculator. It is a block diagram of an original pixel position phase correction | amendment part. It is explanatory drawing which shows the structure of a reading part. It is a comparison figure of the reading result which shows the effect of the present invention.

Explanation of symbols

1 Input Color Image 2 Output Color Image S1 Gray Conversion Step S2 Gray Delay Step S3 Average Value Calculation Step S4 Difference Absolute Value Calculation Step S5 Weighting Calculation Step S6 Color Delay Step S7 Pixel Level Prediction Calculation Step S8 Weighting Average Step S9 Maximum Minimum Density Detection Step S10 Output density limiting step S11 Interpolation position calculation step S12 Original pixel position phase correction step S13 Interpolation pixel output step

Claims (9)

When an image is read by scanning a reading target with a plurality of pixel reading elements arranged in parallel in one column , level values P1, P2, P4, and P5 of continuous pixels are read from the image , and the image is missing. A pixel interpolation method for providing a pixel level value P3 of the interpolation target pixel,
The pixel level value group acquisition means has a pixel level value group assigned to a predetermined number of pixels connected in the first direction from the interpolation target pixel, and a second direction that is opposite to the first direction. Following the pixel level value group acquisition step of acquiring the pixel level value group assigned to the predetermined number of connected pixels,
The pixel level prediction calculation means uses the pixel level prediction formulas expressed by the following equations 1 to 3 as the pixel level prediction calculation means, and the pixel level prediction values FD_L and FD_R in the first direction and the second direction, respectively. A pixel level prediction calculating step and a weighting value calculating means for calculating each pixel level value group by taking a ratio of the square of the deviation on the opposite side to the sum of the squares of the deviations in the first direction and the second direction. A weighting value calculating step for calculating weighting values in the first direction and the second direction using weighting formulas for weighting weighted in the first direction and the second direction, simultaneously or in any order After processing
Interpolation target pixel level value calculating means multiplies the pixel direction predicted value FD_L in the first direction by the weighting value in the first direction, and the pixel direction predicted value FD_R in the second direction in the second direction. A pixel interpolation method comprising: an interpolation target pixel level value calculation step of calculating a pixel level value of the interpolation target pixel using an interpolation target pixel level calculation formula including a calculation of multiplying the weighting value of.
(Equation 1)
When the level value of the pixel is P1> P2 and P4 <P5
Predicted value FD_L = P2- (| P1-P2 |) in the first direction
Prediction value FD_R in the second direction = P4- (| P4-P5 |)
(Equation 2)
When the level value of the pixel is P1 <P2 and P4> P5
Predicted value FD_L = P2 + (| P1-P2 |) in the first direction
Prediction value FD_R = P4 + (| P4-P5 |) in the second direction
(Equation 3)
When the level value of the pixel is other than the relationship of Equation 1 or
Predicted value FD_L = P2 in the first direction
Prediction value FD_R = P4 in the second direction In the pixel level prediction calculation step,
Extracting at least two pixel level values from each of the pixel level value groups;
When the pixel level value increases toward the interpolation target pixel in either the first direction or the second direction, or when the pixel level value decreases toward the interpolation target pixel, the increase rate in each direction Or, while calculating the pixel level value corresponding to the interpolation target pixel from the drop rate, respectively, as the pixel level predicted value,
The pixel interpolation method according to claim 1, wherein in other cases, the pixel level value of a pixel connected to the interpolation target pixel in each direction is set as the respective pixel level predicted value. In the weight value calculating step,
A first average value obtained by averaging the pixel level value group in the first direction, a second average value obtained by averaging the pixel level value group in the second direction, and a predetermined concatenation in both directions from the interpolation target pixel. Calculate the median average value obtained by averaging the pixel level values assigned to the number of pixels,
The first deviation between the first average value and the median average value and the second deviation between the second average value and the median average value are compared, and a larger weight is given to the one with the smaller deviation. The pixel interpolation method described in 1. In the pixel interpolation method, following the interpolation target pixel level value calculating step,
The pixel level value limiting means has at least one of a maximum value and a minimum value from a peripheral pixel level value group assigned to pixels having a predetermined number of peripheral pixels in each of the first direction and the second direction from the interpolation target pixel. 4. The pixel interpolation method according to claim 1, further comprising: a pixel level value limiting step of determining a pixel level value limiting range based on the pixel level value and correcting the pixel level value of the interpolation target pixel to be within the range. . The pixel interpolation method according to any one of claims 1 to 4, wherein the pixel level value is at least one of a hue value, a saturation value, and a brightness value of a pixel. When an image is read by scanning a reading target with a plurality of pixel reading elements arranged in parallel in one column , level values P1, P2, P4, and P5 of continuous pixels are read from the image , and the image is missing. In the pixel interpolation device that gives the pixel level value P3 of the interpolation target pixel,
A pixel level value group assigned to a predetermined number of pixels connected in the first direction from the interpolation target pixel, and a predetermined number of pixels connected in a second direction opposite to the first direction. A pixel level value group obtaining unit for obtaining the obtained pixel level value group;
Pixel level prediction calculation means for calculating the pixel level prediction values in the first direction and the second direction using the pixel level prediction formulas expressed by the following equations 4 to 6 for each pixel level value group;
Each pixel level value group is weighted in the first direction and the second direction by taking the ratio of the square of the deviation on the opposite side to the sum of the squares of the deviations in the first direction and the second direction. Weight value calculating means for calculating the weight values of the first direction and the second direction by inputting to the weight calculation formula;
Interpolation including a calculation of multiplying the predicted pixel level value FD_L in the first direction by the weight value in the first direction and multiplying the predicted pixel level value FD_R in the second direction by the weight value in the second direction A pixel interpolation device comprising: an interpolation target pixel level value calculating unit that calculates a pixel level value of the interpolation target pixel based on a target pixel level calculation formula.
(Equation 4)
When the level value of the pixel is P1> P2 and P4 <P5
Predicted value FD_L = P2- (| P1-P2 |) in the first direction
Prediction value FD_R in the second direction = P4- (| P4-P5 |)
(Equation 5)
When the level value of the pixel is P1 <P2 and P4> P5
Predicted value FD_L = P2 + (| P1-P2 |) in the first direction
Prediction value FD_R = P4 + (| P4-P5 |) in the second direction
(Equation 6)
When the level value of the pixel is other than the relationship of Equation 1 or
Predicted value FD_L = P2 in the first direction
Prediction value FD_R = P4 in the second direction In the pixel interpolation device,
A pixel level value limiting range based on at least one of the maximum value and the minimum value from the peripheral pixel level value group assigned to the pixels having a predetermined number of peripheral pixels in each of the first direction and the second direction from the interpolation target pixel. Decide
The pixel interpolation device according to claim 6, further comprising a pixel level value limiting unit that corrects the pixel level value of the interpolation target pixel calculated by the interpolation target pixel level value calculation unit to be within the range. When an image is read by scanning a reading target with a plurality of pixel reading elements arranged in parallel in one column , level values P1, P2, P4, and P5 of continuous pixels are read from the image , and the image is missing. In the pixel interpolating apparatus that gives the pixel level value P3 of the interpolation target pixel, the image reading apparatus includes:
An image input processing unit that performs an input process of the pixel level value read by the pixel reading element along with scanning and accumulates it as an image;
A pixel level value group assigned to a predetermined number of pixels connected in the first direction from the interpolation target pixel, and a predetermined number of pixels connected in a second direction opposite to the first direction. Pixel level value group acquisition means for acquiring
Pixel level prediction calculation means for calculating each pixel level prediction value in each of the first direction and the second direction using the pixel level prediction formulas represented by Expressions 7 to 9 including the respective pixel level value groups in advance;
Each pixel level value group is weighted in the first direction and the second direction by taking the ratio of the square of the deviation on the opposite side to the sum of the squares of the deviations in the first direction and the second direction. A weighting value calculating means for calculating a weighting value for each of the first direction and the second direction by inputting to the weighting calculation formula;
Interpolation including a calculation of multiplying the predicted pixel level value FD_L in the first direction by the weight value in the first direction and multiplying the predicted pixel level value FD_R in the second direction by the weight value in the second direction A pixel interpolation mechanism comprising each means of interpolation target pixel level value calculation means for calculating a pixel level value of the interpolation target pixel based on a target pixel level calculation formula;
An image reading apparatus comprising: an image output unit that outputs an image including an interpolation target pixel interpolated by the pixel interpolation mechanism.
(Equation 7)
When the level value of the pixel is P1> P2 and P4 <P5
Predicted value FD_L = P2- (| P1-P2 |) in the first direction
Prediction value FD_R in the second direction = P4- (| P4-P5 |)
(Equation 8)
When the level value of the pixel is P1 <P2 and P4> P5
Predicted value FD_L = P2 + (| P1-P2 |) in the first direction
Prediction value FD_R = P4 + (| P4-P5 |) in the second direction
(Equation 9)
When the level value of the pixel is other than the relationship of Equation 1 or Equation 2
Predicted value FD_L = P2 in the first direction
Prediction value FD_R = P4 in the second direction In the image reading apparatus,
The pixel interpolation mechanism is
A pixel level value limiting range based on at least one of the maximum value and the minimum value from the peripheral pixel level value group assigned to the pixels having a predetermined number of peripheral pixels in each of the first direction and the second direction from the interpolation target pixel. Decide
The image reading apparatus according to claim 8, further comprising a pixel level value limiting unit that corrects the pixel level value of the interpolation target pixel calculated by the interpolation target pixel level value calculation unit to be within the range.

Images